On Mon, Nov 21, 2016 at 5:22 AM, Tsunakawa, Takayuki
<tsunakawa.takay@jp.fujitsu.com> wrote:
> From: Tsunakawa, Takayuki/綱川 貴之
>> Thank you, I'll try the read-write test with these settings on the weekend,
>> when my PC is available. I understood that your intention is to avoid being
>> affected by checkpointing and WAL segment creation.
>
> The result looks nice as follows. I took the mean value of three runs.
>
> shared_buffers tps
> 256MB 990
> 512MB 813
> 1GB 1189
> 2GB 2258
> 4GB 5003
> 8GB 5062
>
> "512MB is the largest effective size" seems to be a superstition, although I don't know the reason for the drop at
512MB.
>
It is difficult to say why the performance drops at 512MB, it could be
run-to-run variation. How long have you run each test?
--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com